Messages in this thread | | | Date | Thu, 23 Aug 2012 22:57:27 -0400 | From | Rik van Riel <> | Subject | Re: [RFC][PATCH 0/3] c-state governor changes |
| |
On 08/23/2012 05:54 PM, Arjan van de Ven wrote: > On 8/23/2012 2:11 PM, Rik van Riel wrote: >> This patch set is mostly there to kick off a discussion in time >> for Kernel Summit. When running it on my laptop, with acpi_idle, >> I see a promising change in powertop. > > be careful with acpi_idle... that will remove most of the intermediate C states the platform has, > so the policy engine no longer has good things to chose from.
Interesting, I was not aware of that.
Another thing I did see is that the intel_idle driver selects a target residency time around 4x larger than the exit latency for most c-states, while acpi_idle sticks to 2x by default.
Aiming for a residency time only 2x the exit latency seems excessively aggressive, especially for the deeper c states.
The latency issues with various KVM guests communicating with each other have been observed on various Intel and AMD chipsets, not unique to one vendor.
-- All rights reversed
| |